{"id":"W4385812326","doi":"10.1177/08465371231193716","title":"Comparative Performance of ChatGPT and Bard in a Text-Based Radiology Knowledge Assessment","year":2023,"lang":"en","type":"article","venue":"Canadian Association of Radiologists Journal","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":94,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hamilton Health Sciences; Juravinski Hospital; University of Toronto; McMaster University","funders":"","keywords":"Medicine; Subspecialty; Neuroradiology; Radiology; Nuclear medicine; Medical physics; Pathology; Neurology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001447525,0.00007388533,0.0003658053,0.0005668923,0.00007745577,0.000006212414,0.00005801472,0.0001462509,0.00004290243],"category_scores_gemma":[0.000471731,0.00006900058,0.00004352507,0.0004056465,0.00009257889,0.00005132052,0.000004586867,0.000292553,0.000008084213],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001191163,"about_ca_system_score_gemma":0.002120086,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003019301,"about_ca_topic_score_gemma":0.01680201,"domain_scores_codex":[0.9988543,0.0001905492,0.0004768321,0.0001050676,0.0001083946,0.0002648385],"domain_scores_gemma":[0.9987025,0.000373492,0.0003303683,0.00007378601,0.0003072705,0.0002125942],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000036956,0.00002764481,0.9870203,0.00005640451,0.00003166783,0.000007173402,0.001667513,0.0003182588,0.0002748429,0.0001476425,0.004310393,0.00610118],"study_design_scores_gemma":[0.0002105544,0.0004505684,0.9861992,0.0001105868,0.00001663857,0.00004211427,0.0009712172,0.009703959,0.0005893616,0.000121306,0.001521674,0.00006280884],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9946386,0.0002389359,0.00002669035,0.003535351,0.0003237747,0.000167116,0.00001031909,0.000006797152,0.00105241],"genre_scores_gemma":[0.9988961,0.0004320093,0.0002023898,0.000132404,0.0001170524,0.000008011854,0.00002220414,0.000004715348,0.0001850977],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0137827,"threshold_uncertainty_score":0.9375913,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1115391610351686,"score_gpt":0.4219256449286377,"score_spread":0.3103864838934691,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}